CN109191849A - A kind of traffic congestion Duration Prediction method based on multi-source data feature extraction - Google Patents
A kind of traffic congestion Duration Prediction method based on multi-source data feature extraction Download PDFInfo
- Publication number
- CN109191849A CN109191849A CN201811227002.9A CN201811227002A CN109191849A CN 109191849 A CN109191849 A CN 109191849A CN 201811227002 A CN201811227002 A CN 201811227002A CN 109191849 A CN109191849 A CN 109191849A
- Authority
- CN
- China
- Prior art keywords
- traffic congestion
- data
- vehicle
- section
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000000605 extraction Methods 0.000 title claims abstract description 42
- 238000013528 artificial neural network Methods 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 19
- 239000010410 layer Substances 0.000 claims description 19
- 238000013135 deep learning Methods 0.000 claims description 13
- 239000011229 interlayer Substances 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 7
- 210000005036 nerve Anatomy 0.000 claims description 7
- 230000007787 long-term memory Effects 0.000 claims description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 239000012141 concentrate Substances 0.000 claims description 3
- 238000011144 upstream manufacturing Methods 0.000 claims description 3
- 239000007787 solid Substances 0.000 claims description 2
- 238000003780 insertion Methods 0.000 claims 1
- 230000037431 insertion Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 description 5
- 238000011478 gradient descent method Methods 0.000 description 4
- 238000013480 data collection Methods 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 230000002045 lasting effect Effects 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000013506 data mapping Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012384 transportation and delivery Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
- G08G1/0112—Measuring and analyzing of parameters relative to traffic conditions based on the source of data from the vehicle, e.g. floating car data [FCD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Computation (AREA)
- Marketing (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
Abstract
The traffic congestion Duration Prediction method based on multi-source data feature extraction that this patent discloses a kind of includes three steps: the feature extraction of (one) based on multi-source data；(2) the traffic congestion state prediction based on roadway characteristic；(3) based on the traffic congestion Duration Prediction of multi-source data feature extraction.This patent considers road traffic features used in traffic congestion state prediction process, simultaneously by the output of traffic congestion state prediction technique, foundation one of of the traffic congestion state predicted as congestion time prediction, improves the accuracy that method measures traffic congestion resolution time.
Description
Technical field
The present invention relates to intelligent traffic administration system fields, and in particular to a kind of traffic congestion based on multi-source data feature extraction
Duration Prediction method.
Background technique
With the growth year by year for carrying out China's vehicle guaranteeding organic quantity this year, the traffic of transport need and existing traffic infrastructure
Contradiction between supply capacity is increasingly severe, and traffic congestion is exactly contradictory main between transport need and transportation supplies
It embodies.During traffic management department is managed road traffic state, it cannot be had occurred and that and when traffic congestion
It just taking measures when through constituting economic loss, needing to predict the position that traffic congestion occurs, the information such as time simultaneously take phase in time
Measure is answered to reduce traffic congestion to road network bring adverse effect.Therefore, exploitation one kind can Accurate Prediction traffic gather around
Stifled generation and the traffic forecast method of duration alleviate road traffic congestion and are of great significance for effectively organizing traffic.
Whether the country's patent research of traffic congestion occurs mainly in traffic congestion phenomenon at present, does not consider
The duration of traffic congestion phenomenon.And during actual traffic control, the scale of traffic congestion phenomenon is different, continues
Time is also not quite similar, and the traffic congestion phenomenon of different scales is needed to take different traffic control strategies.Therefore, pre-
The duration for surveying prediction prediction traffic congestion while whether traffic congestion occurs is very necessary.Meanwhile in the current country
In patent, data used in traffic congestion state prediction process be using Floating Car or road fixed detector data this
Class data mapping, the data structure that this results in prediction model used in the prediction process for traffic behavior to obtain is excessively
It is single, it may result in the traffic behavior feature that model cannot react complicated under physical condition well.
Based on this, the present invention is proposed using the data for including fixed detector acquisition, the GPS number of Floating Car real-time delivery
According to multi-source data collection including roadway characteristic data and weather environment data uses the deep learning for considering roadway characteristic
Method, to the method that the duration of traffic congestion event is predicted on the basis of predicting traffic congestion state-event.
Summary of the invention
To solve the above problems, the present invention proposes that a kind of traffic congestion duration based on multi-source data feature extraction is pre-
Survey method, this method is by the way that including the fixed detector being arranged on road, the traffic flow data and road that Floating Car obtains are special
Data are levied, the multi-source data including weather data carries out traffic characteristic, the extraction of roadway characteristic and weather characteristics, and use is based on
The method of the deep learning of multi-source data feature extraction carries out the duration of traffic congestion generating state and traffic congestion
Prediction.
Data source of the present invention includes fixed detector acquisition data on road, the GPS data that Floating Car uploads, road spy
Levy data and weather data.Wherein: the data that fixed detector obtains include vehicle identification number, position and pass through time data；It is floating
Motor-car GPS data includes dynamic vehicle identification number, location and time data；Roadway characteristic data include each section on road network
There are situations for position number of track-lines and bottleneck；Weather data includes under obtaining data phase in the same time with fixed detector and Floating Car
Rainfall product data.
Traffic congestion Duration Prediction method proposed by the present invention based on multi-source data feature extraction includes three steps
It is rapid: the feature extraction of (one) based on multi-source data；(2) the traffic congestion state prediction based on roadway characteristic；(3) it is based on multi-source
The traffic congestion Duration Prediction that data characteristics is extracted.
(1) based on the feature extraction of multi-source data
It includes fixed detector acquisition data, the GPS data that Floating Car uploads, roadway characteristic number on road that the present invention, which uses,
According to and weather data including multiple data sources data, these data sources provide data in structure and meaning may
It is not quite similar, therefore before carrying out traffic congestion state and the prediction of duration using the multi-source data got, needs
Feature extraction is carried out to multi-source data, keeps mode input data structure unified, clear.The specific steps of which are as follows:
Step 1: obtain fixed detector data, floating car data, and it be stored in respectively fixed detector data set with
Floating car data collection.
Step 2: extracting floating car data and concentrate vehicle data, according to wherein vehicle identification number, vehicle position information data will
It is matched with the vehicle data that fixed detector obtains, and is stored in data set.
Step 3: the data that fixed detector obtains being subjected to traffic characteristic in conjunction with the GPS data that Floating Car uploads and are mentioned
It takes, method particularly includes: it extracts fixed detector and detects the time that vehicle passes through, it is inserted into according to chronological order and is floated
In the GPS data that motor-car uploads, more accurate track of vehicle data are obtained.
Step 4: according to the track of vehicle drivingAnd vehicle passes through the specific time of each point on trackWherein i=
1,2,3 ... be car number, j=1,2,3 ... be track of vehicle point number, available vehicle is any two on road
Travel speed between a tracing pointVehicle average overall travel speed of the vehicle i between tracing point j and j+1 is represented, i.e.,It can
It indicates are as follows:
Step 5: according to number of track-lines data in roadway characteristic data, the position that number of track-lines is changed is divided as section
Node, and be according to dividing road section, specific practice according to fixed range are as follows: number of track-lines on road is changed or
There are the places of intersection as road section division boundary, and the too long section of link length is divided into along fixed range
Segment.
Step 6: using set time length Δ t as time interval, vehicle being divided into the time of Δ t by the time in section
Section, and the tracing point time is passed through according to vehicleThe track of vehicle time is divided in different time sections, and is drawn in conjunction with step 5
Track of vehicle is divided to different sections of highway by the section divided, it is possible thereby to driving status of the vehicle at any tracing point is included into solid
Determine in the fixed time period of section.
Step 7: travel speed of the vehicle between the continuous tracing point j and j+1 of any two that step 4 is obtained?
Time and Spatial Dimension are averaged to obtain road average-speed in any time period, i.e., divide it according to the section that step 5 divides
It separates, and calculates the average speed on sectionWherein k indicates that step 5 divides obtained section number, and subscript t indicates average
SpeedBelong in the period marked as t, then road average-speedFor；
Wherein, m is the vehicle fleet passed through on the k of section in time period t, and n is the vehicle passed through on the k of section in time period t
Track points, N indicates that k upper all tracing points by vehicle in section are total in time period t, i.e.,
Step 8: concentrating the road obtained in each period in every a road section from roadway characteristic data set and weather data
And weather characteristics data, including section number of lanes nl, whether there is bottleneck 0/1 with upstream and downstream lane, wherein 0 indicates not
Indicate that there are traffic bottlenecks, rainfall p there are traffic bottlenecks, 1, and the road average-speed data for combining step 7 to obtainStructure
Build the input of deep neural network.
(2) the traffic congestion state prediction based on roadway characteristic
For referring to that basis is to be to traffic congestion before the Duration Prediction of the traffic congestion phenomenon occurred on road
The prediction of no generation, i.e., to the prediction of traffic congestion state.The shot and long term memory network for considering roadway characteristic is used in the present invention
The deep learning of LSTM predicts traffic congestion state.
Consider that the deep neural network of the LSTM of roadway characteristic, network inputs include one and include used in the present invention
Road average-speed dataTime segment number t, section number of lanes nl, if there are bottleneck 0/1 and the vector of rainfall p,
That is vectorThe wherein amount that bottleneck is 0 or 1, indicates whether there are bottleneck, when
It when bottleneck is 0, indicates that bottleneck is not present, indicates that there are bottlenecks when bottleneck is 1.The present invention passes through to road
The extraction of feature, and it is added into the input of neural network, to consider influence of the roadway characteristic to traffic congestion state, to obtain
Obtain more accurate precision of prediction.It can be by remembering on time dimension using the deep neural network for considering roadway characteristic
Transmitting, realizes connection of the input in time dimension of data, i.e. model can learn the temporal continuation relationship of traffic data,
To obtain compared to the more accurate prediction result of conventional traffic prediction algorithm.
The input of deep learning algorithm is vector in the present inventionIn training for neural network
Cheng Zhong, it is desirable to provide model has a training data of label, and traffic congestion is divided into 5 states in the present invention, and with number 0,1,
2,3,4 indicate, wherein 0 indicates that congestion phenomenon is not present in the coast is clear, 4 indicate that road Severe blockage, vehicle can not pass through.This hair
Use road average-speed as partitioning standards the foundation that traffic congestion state divides in bright, with specific reference to the practical item of road
Part determines the boundary for dividing different traffic congestion states.By increasing label to data, model in available training process
InputAnd model label y.Input is added in model in the training process and is calculated currently
Prediction output under parameterAnd loss function is combined to calculate penalty values, wherein loss function is cross entropy
Due to considering that the deep learning model of the LSTM of roadway characteristic has a large amount of calculating, model on time dimension
Depth should not be too large, and export structure of the three layers of LSTM nervous layer as model is used in the present invention, and interlayer LSTM unit uses
Tanh function is as activation primitive.Meanwhile for the generalization ability of strength neural network, reduce what it occurred in the training process
Over-fitting increases the Dropout layers of method as model regularization in LSTM nerve interlayer.Further, due to LSTM
Model needs largely to calculate on time dimension cost, and the training effect of traditional stochastic gradient descent method is generally poor, therefore
Model is trained using Adam gradient descent algorithm.Further, the present invention in traffic congestion state be predicted as it is more
Classification problem, therefore increase sigmoid function as model output layer activation primitive in model output layer.
(3) based on the traffic congestion Duration Prediction of multi-source data feature extraction
Traffic congestion Duration Prediction method proposed by the present invention based on multi-source data feature extraction gathers around traffic
The prediction of stifled duration, which needs to export based on the prediction in the present invention partially (two) for traffic congestion state, to be carried out.
The traffic congestion state obtained first to prediction in part (two) judges, when the traffic congestion shape that prediction obtains
When state is 0, it is believed that any type of traffic congestion is not present, vehicle is traveled freely on road, therefore is not gathered around to its traffic
The stifled duration is predicted.When traffic congestion state is 1,2,3,4, illustrate that different degrees of traffic has occurred on road gathers around
Stifled phenomenon, needs to predict its traffic congestion duration at this time, and since traffic congestion degree is different, congestion is lasting
Time is also not exactly the same, i.e., traffic congestion degree is to influence the key factor of traffic congestion duration.Based on this, make
When with the deep learning method based on multi-source data feature extraction to traffic congestion Duration Prediction, need in part (two)
Increase traffic congestion state on the basis of the input of model, i.e. the traffic congestion duration based on multi-source data feature extraction is pre-
The input of survey method isWherein state indicates that the state of traffic congestion, value are desirable
It is 1,2,3,4.The output adjustment of model is traffic congestion duration tc, i.e. y=tc.In the training process there is still a need for for number
According to label is increased, then the traffic congestion time actually measured when traffic congestion occurring on road is used to predict mould as traffic congestion
The output label of type.
It is contacted since traffic congestion duration and traffic flow parameter exist in the continuity of time, traffic is gathered around
The prediction of stifled duration still use can well between learning data the LSTM neural network of time relationship as deep learning side
The basic framework of method.3 layers of LSTM neural network are used for the prediction model of traffic congestion duration, interlayer LSTM unit makes
Use tanh function as model activation primitive.Input is added in model in the training process, Current Situation of Neural Network ginseng is calculated
Prediction output under said conditionsAnd loss function is combined to calculate penalty values, loss function is cross entropyMeanwhile it is similar with LSTM neural network in part (two), in order to enhance nerve
The generalization ability of network uses and increases Dropout layers in LSTM nerve interlayer and reduce it as the method for model regularization and instructing
The over-fitting occurred during practicing.Further, there are a large amount of calculating on time dimension due to LSTM model, it is traditional
The training effect of stochastic gradient descent method is generally poor, therefore is trained using Adam gradient descent algorithm to model.
The present invention has the advantages that
(1) traffic flow data and roadway characteristic number obtained using the fixed detector including being arranged on road, Floating Car
According to the multi-source data including weather data carries out feature extraction, and the feature after extraction as traffic congestion state and is continued
The input of time forecasting methods can obtain traffic and the roadway characteristic of degree of precision by the extraction of feature.
(2) using the LSTM deep learning method for considering roadway characteristic, traffic is added in bottleneck information intrinsic on road
In the prediction of congestion status, it may be implemented to combine accurate prediction of the real road condition to traffic congestion state, improve result
Confidence level.
(3) in the traffic congestion Duration Prediction method proposed by the present invention based on multi-source data feature extraction, not only
Consider road traffic features used in traffic congestion state prediction process, while by the defeated of traffic congestion state prediction technique
Out, that is, foundation one of of the traffic congestion state predicted as congestion time prediction, when improving method to traffic congestion dissipation
Between the accuracy measured, make this method can be with traffic conditions complicated under the conditions of good conformity real road.
Detailed description of the invention
Fig. 1 is the traffic congestion Duration Prediction method flow diagram of the invention based on multi-source data feature extraction
Specific embodiment
Below with reference to specific example and attached drawing, the invention will be further described
The present invention proposes that a kind of traffic congestion Duration Prediction method based on multi-source data feature extraction, this method are logical
Cross the traffic flow data obtained to the fixed detector including being arranged on road, Floating Car and roadway characteristic data, weather data
Multi-source data inside carries out traffic characteristic, the extraction of roadway characteristic and weather characteristics, and mentions using based on multi-source data feature
The method of the deep learning taken predicts the duration of traffic congestion generating state and traffic congestion.
Data source of the present invention includes fixed detector acquisition data on road, the GPS data that Floating Car uploads, road spy
Levy data and weather data.Wherein: the data that fixed detector obtains include vehicle identification number, position and pass through time data；It is floating
Motor-car GPS data includes dynamic vehicle identification number, location and time data；Roadway characteristic data include each section on road network
There are situations for position number of track-lines and bottleneck；Weather data includes under obtaining data phase in the same time with fixed detector and Floating Car
Rainfall product data.
Traffic congestion Duration Prediction method proposed by the present invention based on multi-source data feature extraction includes three steps
It is rapid: the feature extraction of (one) based on multi-source data；(2) the traffic congestion state prediction based on roadway characteristic；(3) it is based on multi-source
The traffic congestion Duration Prediction that data characteristics is extracted.
(1) based on the feature extraction of multi-source data
It includes fixed detector acquisition data, the GPS data that Floating Car uploads, roadway characteristic number on road that the present invention, which uses,
According to and weather data including multiple data sources data, these data sources provide data in structure and meaning may
It is not quite similar, therefore before carrying out traffic congestion state and the prediction of duration using the multi-source data got, needs
Feature extraction is carried out to multi-source data, keeps mode input data structure unified, clear.The specific steps of which are as follows:
Step 1: obtain fixed detector data, floating car data, and it be stored in respectively fixed detector data set with
Floating car data collection.
Step 2: extracting floating car data and concentrate vehicle data, according to wherein vehicle identification number, vehicle position information data will
It is matched with the vehicle data that fixed detector obtains, and is stored in data set.
Step 3: the data that fixed detector obtains being subjected to traffic characteristic in conjunction with the GPS data that Floating Car uploads and are mentioned
It takes, method particularly includes: it extracts fixed detector and detects the time that vehicle passes through, it is inserted into according to chronological order and is floated
In the GPS data that motor-car uploads, more accurate track of vehicle data are obtained.
Step 4: according to the track of vehicle drivingAnd vehicle passes through the specific time of each point on trackWherein i=
1,2,3 ... be car number, j=1,2,3 ... be track of vehicle point number, available vehicle is any two on road
Travel speed between a tracing pointVehicle average overall travel speed of the vehicle i between tracing point j and j+1 is represented, i.e.,It can
It indicates are as follows:
Step 5: according to number of track-lines data in roadway characteristic data, the position that number of track-lines is changed is divided as section
Node, and be according to dividing road section, specific practice according to fixed range are as follows: number of track-lines on road is changed or
There are the places of intersection as road section division boundary, and the too long section of link length is divided into along fixed range
Segment.
Step 6: using set time length 5min as time interval, i.e. Δ t=5min draws vehicle by the time in section
It is divided into the period of 5min, and the tracing point time is passed through according to vehicleThe track of vehicle time is divided in different time sections,
And track of vehicle is divided to different sections of highway by the section for combining step 5 to divide, it is possible thereby to by vehicle at any tracing point
Driving status is included into fixed section fixed time period.
Step 7: travel speed of the vehicle between the continuous tracing point j and j+1 of any two that step 4 is obtained?
Time and Spatial Dimension are averaged to obtain road average-speed in any time period, i.e., divide it according to the section that step 5 divides
It separates, and calculates the average speed on sectionWherein k indicates that step 5 divides obtained section number, and subscript t indicates average
SpeedBelong in the period marked as t, then road average-speedAre as follows:
Wherein, m is the vehicle fleet passed through on the k of section in time period t, and n is the vehicle passed through on the k of section in time period t
Track points, N indicates that k upper all tracing points by vehicle in section are total in time period t, i.e.,Such as at certain
In the time interval of a 5min, 3 are shared by the vehicle on section, the travel speed on section between tracing point is respectively
First car Second car
Third vehicleThen N=3+3+4=10,I.e. in this 5min time interval, this section
Upper average speed is 20.2m/s.
Step 8: concentrating the road obtained in each period in every a road section from roadway characteristic data set and weather data
And weather characteristics data, including section number of lanes nl, whether there is bottleneck 0/1 with upstream and downstream lane, wherein 0 indicates not
Indicate that there are traffic bottlenecks, rainfall p there are traffic bottlenecks, 1, and the road average-speed data for combining step 7 to obtainStructure
Build the input of deep neural network.
(2) the traffic congestion state prediction based on roadway characteristic
For referring to that basis is to be to traffic congestion before the Duration Prediction of the traffic congestion phenomenon occurred on road
The prediction of no generation, i.e., to the prediction of traffic congestion state.The shot and long term memory network for considering roadway characteristic is used in the present invention
The deep learning of LSTM predicts traffic congestion state.
Consider that the deep neural network of the LSTM of roadway characteristic, network inputs include one and include used in the present invention
Road average-speed dataTime segment number t, section number of lanes nl, if there are bottleneck 0/1 and the vector of rainfall p,
That is vectorThe wherein amount that bottleneck is 0 or 1, indicates whether there are bottleneck, when
It when bottleneck is 0, indicates that bottleneck is not present, indicates that there are bottlenecks when bottleneck is 1.The present invention passes through to road
The extraction of feature, and it is added into the input of neural network, to consider influence of the roadway characteristic to traffic congestion state, to obtain
Obtain more accurate precision of prediction.It can be by remembering on time dimension using the deep neural network for considering roadway characteristic
Transmitting, realizes connection of the input in time dimension of data, i.e. model can learn the temporal continuation relationship of traffic data,
To obtain compared to the more accurate prediction result of conventional traffic prediction algorithm.
The input of deep learning algorithm is vector in the present inventionIn training for neural network
Cheng Zhong, it is desirable to provide model has a training data of label, and traffic congestion is divided into 5 states in the present invention, and with number 0,1,
2,3,4 indicate, wherein 0 indicates that congestion phenomenon is not present in the coast is clear, 4 indicate that road Severe blockage, vehicle can not pass through.This hair
Use road average-speed as partitioning standards the foundation that traffic congestion state divides in bright, with specific reference to the practical item of road
Part determines the boundary for dividing different traffic congestion states.Using Beijing's provincial standard, " urban road traffic congestion is commented in this example
Valence index system " in the criteria for classifying of traffic congestion state is determined, as shown in table 1.
| Grade | 1 | 2 | 3 | 4 | 5 |
| Through street | >65 | (50,65] | (35,50] | (20,35] | ≤20 |
| Trunk roads | >45 | (35,45] | (25,35] | (15,25] | ≤15 |
| Secondary distributor road | >35 | (25,35] | (15,25] | (10,15] | ≤10 |
| Branch | >35 | (25,35] | (15,25] | (10,15] | ≤10 |
Table 1
By increasing label, the input of available model to dataAnd model label
y.The prediction being calculated under parameter current is added in model in input in the training process to exportAnd combine loss function meter
Penalty values are calculated, wherein loss function is cross entropy
Due to considering that the deep learning model of the LSTM of roadway characteristic has a large amount of calculating, model on time dimension
Depth should not be too large, and export structure of the three layers of LSTM nervous layer as model is used in the present invention, and interlayer LSTM unit uses
Tanh function is as activation primitive.Meanwhile for the generalization ability of strength neural network, reduce what it occurred in the training process
Over-fitting increases the Dropout layers of method as model regularization in LSTM nerve interlayer.Further, due to LSTM
Model needs largely to calculate on time dimension cost, and the training effect of traditional stochastic gradient descent method is generally poor, therefore
Model is trained using Adam gradient descent algorithm.Further, the present invention in traffic congestion state be predicted as it is more
Classification problem, therefore increase sigmoid function as model output layer activation primitive in model output layer.
(3) based on the traffic congestion Duration Prediction of multi-source data feature extraction
Traffic congestion Duration Prediction method proposed by the present invention based on multi-source data feature extraction gathers around traffic
The prediction of stifled duration, which needs to export based on the prediction in the present invention partially (two) for traffic congestion state, to be carried out.
The traffic congestion state obtained first to prediction in part (two) judges, when the traffic congestion shape that prediction obtains
When state is 0, it is believed that any type of traffic congestion is not present, vehicle is traveled freely on road, therefore is not gathered around to its traffic
The stifled duration is predicted.When traffic congestion state is 1,2,3,4, illustrate that different degrees of traffic has occurred on road gathers around
Stifled phenomenon, needs to predict its traffic congestion duration at this time, and since traffic congestion degree is different, congestion is lasting
Time is also not exactly the same, i.e., traffic congestion degree is to influence the key factor of traffic congestion duration.Based on this, make
When with the deep learning method based on multi-source data feature extraction to traffic congestion Duration Prediction, need in part (two)
Increase traffic congestion state on the basis of the input of model, i.e. the traffic congestion duration based on multi-source data feature extraction is pre-
The input of survey method isWherein state indicates that the state of traffic congestion, value are desirable
It is 1,2,3,4.The output adjustment of model is traffic congestion duration tc, i.e. y=tc.In the training process there is still a need for for number
According to label is increased, then the traffic congestion time actually measured when traffic congestion occurring on road is used to predict mould as traffic congestion
The output label of type.
It is contacted since traffic congestion duration and traffic flow parameter exist in the continuity of time, traffic is gathered around
The prediction of stifled duration still use can well between learning data the LSTM neural network of time relationship as deep learning side
The basic framework of method.3 layers of LSTM neural network are used for the prediction model of traffic congestion duration, interlayer LSTM unit makes
Use tanh function as model activation primitive.Input is added in model in the training process, Current Situation of Neural Network ginseng is calculated
Prediction output under said conditionsAnd loss function is combined to calculate penalty values, loss function is cross entropyMeanwhile it is similar with LSTM neural network in part (two), in order to enhance nerve
The generalization ability of network uses and increases Dropout layers in LSTM nerve interlayer and reduce it as the method for model regularization and instructing
The over-fitting occurred during practicing.Further, there are a large amount of calculating on time dimension due to LSTM model, it is traditional
The training effect of stochastic gradient descent method is generally poor, therefore is trained using Adam gradient descent algorithm to model.
Claims (4)
1. a kind of traffic congestion Duration Prediction method based on multi-source data feature extraction, which is characterized in that the method
Include the following steps:
Step 1: the feature extraction based on more former data comprising following sub-step；
S101: fixed detector data, floating car data are obtained, and it is stored in fixed detector data set and Floating Car respectively
Data set.S102: it extracts floating car data and concentrates vehicle data, according to wherein vehicle identification number, vehicle position information data, by it
The vehicle data obtained with fixed detector is matched, and is stored in data set；S103: the data that fixed detector is obtained
Traffic characteristic extraction is carried out in conjunction with the GPS data that Floating Car uploads, method particularly includes: it extracts fixed detector and detects vehicle
Time passed through in the GPS data for uploading it according to chronological order insertion Floating Car, obtains track of vehicle data；
S104: according to the track of vehicle drivingAnd vehicle passes through the specific time of each point on trackWherein i=1,2,3 ...
For car number, j=1,2,3 ... it is the number of track of vehicle point, obtains vehicle on road between any two tracing point
Travel speedVehicle average overall travel speed of the vehicle i between tracing point j and j+1 is represented, i.e.,It indicates are as follows:S105: according to number of track-lines data in roadway characteristic data, the position that number of track-lines is changed is as section
The node of division, and be according to division road section, specific practice are as follows: change number of track-lines on road according to fixed range
Become or there are the place of intersection as road section divide boundary, and by the too long section of link length along fixed range draw
It is divided into segment；S106: using set time length Δ t as time interval, vehicle is divided into the time of Δ t by the time in section
Section, and the tracing point time is passed through according to vehicleThe track of vehicle time is divided in different time sections, and is drawn in conjunction with step 5
Track of vehicle is divided to different sections of highway by the section divided, it is possible thereby to driving status of the vehicle at any tracing point is included into solid
Determine in the fixed time period of section；S107: row of the vehicle between the continuous tracing point j and j+1 of any two that S104 is obtained
Sail speedIt is averaged to obtain road average-speed in any time period in time and Spatial Dimension, i.e., draws it according to step 5
The section divided separates, and calculates the average speed on sectionWherein k indicates the section number that S105 is divided, subscript t
Indicate average speedBelong in the period marked as t, then road average-speedFor；Wherein, m is the time
The vehicle fleet passed through on the k of section in section t, n are the track points of the vehicle passed through on the k of section in time period t, and N indicates the time
All tracing point sums by vehicle on the k of section in section t, i.e.,S108: from roadway characteristic data set and weather
The road and weather characteristics data in each period in every a road section are obtained in data set, including section number of lanes
nl, whether there is bottleneck 0/1 with upstream and downstream lane, wherein 0 indicates that traffic bottlenecks are not present, 1 indicates that there are traffic bottlenecks, rainfalls
Measure p, and the road average-speed data for combining step 7 to obtainConstruct the input of deep neural network；
Step 2: the traffic congestion state based on roadway characteristic is predicted
Middle use considers that the deep learning of the shot and long term memory network LSTM of roadway characteristic predicts traffic congestion state,
It includes road average-speed data that network inputs, which include one,Time segment number t, section number of lanes nl, if there are bottles
The vector of neck 0/1 and rainfall p, i.e. vectorThe wherein amount that bottleneck is 0 or 1, expression is
No there are bottlenecks, when bottleneck is 0, indicate that bottleneck is not present, indicate that there are bottlenecks when bottleneck is 1；Note
Recall having in the training data of label for network LSTM, traffic congestion be divided into 5 states, and indicate with number 0,1,2,3,4,
In 0 indicate that congestion phenomenon is not present in the coast is clear, 4 indicate road Severe blockages, and vehicle can not pass through；It in the training process will be defeated
Enter the prediction output for being added in model and being calculated under parameter currentAnd loss function is combined to calculate penalty values；
Step 3: the traffic congestion Duration Prediction based on multi-source data feature extraction
The traffic congestion state obtained first to prediction in part (two) judges, when the traffic congestion state that prediction obtains is
When 0, its traffic congestion duration is not predicted；When traffic congestion state is 1,2,3,4, its traffic congestion is continued
Time is predicted；The prediction model that the traffic congestion Duration Prediction uses is the input of model in part (two)
On the basis of increase traffic congestion state, the i.e. input of the traffic congestion Duration Prediction method based on multi-source data feature extraction
ForWherein state indicates that the state of traffic congestion, value are taken as 1,2,3,4；Model
Output be traffic congestion duration tc, i.e. y=tc；When during model training using traffic congestion occurs on road
Output label of the actually measured traffic congestion time as traffic congestion prediction model.
2. a kind of traffic congestion Duration Prediction method based on multi-source data feature extraction according to claim 1,
It is characterized in that, loss function is cross entropy in shot and long term memory network LSTMExport structure using three layers of LSTM nervous layer as model, interlayer LSTM unit
Use tanh function as activation primitive；Increase the Dropout layers of method as model regularization in LSTM nerve interlayer.
3. a kind of traffic congestion Duration Prediction method based on multi-source data feature extraction according to claim 1,
It is characterized in that, being trained using Adam gradient descent algorithm to model in shot and long term memory network LSTM.
4. a kind of traffic congestion Duration Prediction method based on multi-source data feature extraction according to claim 1,
It is characterized in that, increasing sigmoid function in shot and long term memory network LSTM as model output layer in model output layer and swashing
Function living.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811227002.9A CN109191849B (en) | 2018-10-22 | 2018-10-22 | Traffic jam duration prediction method based on multi-source data feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811227002.9A CN109191849B (en) | 2018-10-22 | 2018-10-22 | Traffic jam duration prediction method based on multi-source data feature extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109191849A true CN109191849A (en) | 2019-01-11 |
| CN109191849B CN109191849B (en) | 2020-10-09 |
Family
ID=64945987
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811227002.9A Expired - Fee Related CN109191849B (en) | 2018-10-22 | 2018-10-22 | Traffic jam duration prediction method based on multi-source data feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109191849B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993356A (en) * | 2019-03-25 | 2019-07-09 | 广东工业大学 | A kind of traffic optimization system and method based on XGBOOST |
| CN110335465A (en) * | 2019-07-10 | 2019-10-15 | 北京维联众诚科技有限公司 | Traffic jam detection method and system in monitor video based on AI deep learning |
| CN110827540A (en) * | 2019-11-04 | 2020-02-21 | 黄传明 | Motor vehicle movement mode recognition method and system based on multi-mode data fusion |
| CN110889427A (en) * | 2019-10-15 | 2020-03-17 | 同济大学 | Congestion traffic flow traceability analysis method |
| CN110942642A (en) * | 2019-11-20 | 2020-03-31 | 中科视元科技(杭州)有限公司 | Video-based traffic slow-driving detection method and system |
| CN111462485A (en) * | 2020-03-31 | 2020-07-28 | 电子科技大学 | Traffic intersection congestion prediction method based on machine learning |
| CN111680745A (en) * | 2020-06-08 | 2020-09-18 | 青岛大学 | Burst congestion judging method and system based on multi-source traffic big data fusion |
| CN112598907A (en) * | 2020-12-17 | 2021-04-02 | 东风商用车有限公司 | Congestion prediction method, device, equipment and readable storage medium |
| CN113160570A (en) * | 2021-05-27 | 2021-07-23 | 长春理工大学 | Traffic jam prediction method and system |
| CN118469207A (en) * | 2024-05-17 | 2024-08-09 | 苏州蓝水软件开发股份有限公司 | Method and system for scheduling bus transportation tasks based on cloud management |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200090A (en) * | 2014-08-27 | 2014-12-10 | 百度在线网络技术（北京）有限公司 | Prediction method and method based on multi-source heterogeneous data |
| CN105702041A (en) * | 2016-04-21 | 2016-06-22 | 东南大学 | Highway multisource data fusion state estimation system based on neural network and method thereof |
| CN106971536A (en) * | 2017-04-17 | 2017-07-21 | 武汉云砥信息科技有限公司 | It is a kind of to merge Floating Car and the highway congestion status detection method of trackside video |
| CN107967803A (en) * | 2017-11-17 | 2018-04-27 | 东南大学 | Traffic congestion Forecasting Methodology based on multi-source data and variable-weight combined forecasting model |
| CN108629973A (en) * | 2018-05-11 | 2018-10-09 | 四川九洲视讯科技有限责任公司 | Road section traffic volume congestion index computational methods based on fixed test equipment |
-
2018
- 2018-10-22 CN CN201811227002.9A patent/CN109191849B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200090A (en) * | 2014-08-27 | 2014-12-10 | 百度在线网络技术（北京）有限公司 | Prediction method and method based on multi-source heterogeneous data |
| CN105702041A (en) * | 2016-04-21 | 2016-06-22 | 东南大学 | Highway multisource data fusion state estimation system based on neural network and method thereof |
| CN106971536A (en) * | 2017-04-17 | 2017-07-21 | 武汉云砥信息科技有限公司 | It is a kind of to merge Floating Car and the highway congestion status detection method of trackside video |
| CN107967803A (en) * | 2017-11-17 | 2018-04-27 | 东南大学 | Traffic congestion Forecasting Methodology based on multi-source data and variable-weight combined forecasting model |
| CN108629973A (en) * | 2018-05-11 | 2018-10-09 | 四川九洲视讯科技有限责任公司 | Road section traffic volume congestion index computational methods based on fixed test equipment |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109993356A (en) * | 2019-03-25 | 2019-07-09 | 广东工业大学 | A kind of traffic optimization system and method based on XGBOOST |
| CN110335465A (en) * | 2019-07-10 | 2019-10-15 | 北京维联众诚科技有限公司 | Traffic jam detection method and system in monitor video based on AI deep learning |
| CN110889427A (en) * | 2019-10-15 | 2020-03-17 | 同济大学 | Congestion traffic flow traceability analysis method |
| CN110827540B (en) * | 2019-11-04 | 2021-03-12 | 黄传明 | Motor vehicle movement mode recognition method and system based on multi-mode data fusion |
| CN110827540A (en) * | 2019-11-04 | 2020-02-21 | 黄传明 | Motor vehicle movement mode recognition method and system based on multi-mode data fusion |
| CN110942642A (en) * | 2019-11-20 | 2020-03-31 | 中科视元科技(杭州)有限公司 | Video-based traffic slow-driving detection method and system |
| CN110942642B (en) * | 2019-11-20 | 2021-01-19 | 中科视元科技(杭州)有限公司 | Video-based traffic slow-driving detection method and system |
| CN111462485A (en) * | 2020-03-31 | 2020-07-28 | 电子科技大学 | Traffic intersection congestion prediction method based on machine learning |
| CN111680745A (en) * | 2020-06-08 | 2020-09-18 | 青岛大学 | Burst congestion judging method and system based on multi-source traffic big data fusion |
| CN111680745B (en) * | 2020-06-08 | 2021-03-16 | 青岛大学 | Burst congestion judging method and system based on multi-source traffic big data fusion |
| CN112598907A (en) * | 2020-12-17 | 2021-04-02 | 东风商用车有限公司 | Congestion prediction method, device, equipment and readable storage medium |
| CN112598907B (en) * | 2020-12-17 | 2021-12-07 | 东风商用车有限公司 | Congestion prediction method, device, equipment and readable storage medium |
| CN113160570A (en) * | 2021-05-27 | 2021-07-23 | 长春理工大学 | Traffic jam prediction method and system |
| CN118469207A (en) * | 2024-05-17 | 2024-08-09 | 苏州蓝水软件开发股份有限公司 | Method and system for scheduling bus transportation tasks based on cloud management |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109191849B (en) | 2020-10-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109191849A (en) | A kind of traffic congestion Duration Prediction method based on multi-source data feature extraction | |
| CN104778834B (en) | Urban road traffic jam judging method based on vehicle GPS data | |
| Fang et al. | FTPG: A fine-grained traffic prediction method with graph attention network using big trace data | |
| Zhao et al. | Truck traffic speed prediction under non-recurrent congestion: Based on optimized deep learning algorithms and GPS data | |
| CN105825669B (en) | A kind of system and method for identifying urban expressway traffic bottleneck | |
| CN109670277A (en) | A kind of travel time prediction method based on multi-modal data fusion and multiple-model integration | |
| CN102592447B (en) | Method for judging road traffic state of regional road network based on fuzzy c means (FCM) | |
| CN102521965B (en) | Effect evaluation method of traffic demand management measures based on identification data of license plates | |
| CN104050237B (en) | A kind of road mapping method and system | |
| CN106327871B (en) | A kind of crowded prediction technique of highway of fusion historical data and reservation data | |
| WO2023109720A1 (en) | Urban public transportation planning method | |
| CN102509454B (en) | Road state merging method based on floating car data (FCD) and earth magnetism detector | |
| CN102722986B (en) | Urban road network traffic control subarea dynamic partitioning method | |
| CN107248283A (en) | A kind of urban area road network evaluation of running status method of consideration section criticality | |
| CN106384509A (en) | Urban road driving time distribution estimation method considering taxi operation states | |
| CN101739822B (en) | Sensor network configuring method for regional traffic state acquisition | |
| CN109711640A (en) | A kind of Short-time Traffic Flow Forecasting Methods based on fuzzy C-mean algorithm magnitude of traffic flow cluster and error feedback convolutional neural networks | |
| CN104809112A (en) | Method for comprehensively evaluating urban public transportation development level based on multiple data | |
| CN105374204B (en) | A kind of method that urban highway traffic detector is layouted | |
| CN104715610A (en) | Urban traffic-oriented traffic index calculation method | |
| CN110363985A (en) | A kind of traffic data analyzing method, apparatus, storage medium and equipment | |
| CN107085939B (en) | A kind of highway VMS layout optimization method divided based on road network grade | |
| CN108108859A (en) | A kind of traffic administration duties optimization method based on big data analysis | |
| CN104464280B (en) | Vehicle advance expenditure prediction method and system | |
| CN103049829A (en) | Integrated fusion method of urban and rural passenger line network and hub station |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201009 Termination date: 20211022 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |